The GCP connector is in Beta. Its scanners are verified against emulated GCP APIs; they have not yet been validated on a live customer account.
Today the connector authenticates with a service-account key: you create a service account in your project, grant it read-only roles, and upload a JSON key, which we store encrypted at rest. Workload Identity Federation — no key at all — is where we want to get to, and is not built yet. You can delete the key in your project at any time.
1. Compute
- Compute Engine VMs — terminated instances, public IPs, missing deletion protection, preemptible configuration.
- Cloud Run / Cloud Functions — public invokers (
allUsers), end-of-life runtimes, no minimum instances, no VPC connector, unrestricted ingress. - GKE — covered in the Kubernetes post.
2. Data
- Cloud Storage — any role granted to
allUsersorallAuthenticatedUsers, uniform bucket-level access off, no retention policy, versioning off, Google-managed rather than customer-managed encryption. - Cloud SQL — backups, high availability, SSL enforcement, public IP, engine end-of-life.
- BigQuery — public or externally shared datasets, tables with no expiration, sensitive-looking column names with no policy tag. We read schemas, never table data.
- Secret Manager — rotation configuration, version sprawl, encryption key type. Never values: the scanner does not call the API that returns them.
- Pub/Sub and Cloud Logging — configuration checks, including log-bucket retention and disabled sinks.
3. Identity
- Primitive roles (Owner/Editor/Viewer) bound at the project level, and public or external members.
- User-managed service-account keys — their presence, and their age.
4. Security Command Center and Service Health
If SCC is enabled and you grant access, we pull its active findings. This one is opt-in. Personalized Service Health: active Google Cloud incidents that touch the project.
5. Cost
Google has no per-project spend API, so spend comes from your own Cloud Billing export to BigQuery. If you have set one up and given the service account read access to it, a nightly check flags a day well above the week before it and budgets reaching 75, 90 and 100%. It is off unless it is switched on for your account, because the query is billed to your project.
What the GCP connector does NOT do
- Never reads Cloud Storage object contents, BigQuery table data, Cloud SQL rows, or Secret Manager values.
- Never writes. Every role we ask for is a read-only viewer role; the exact list is in the onboarding wizard.
- Never enables APIs on your behalf. A check whose API is disabled reports as unavailable.
Service Health incidents are read every 5 minutes on a paid plan (every 15 during a trial). Security Command Center findings are not read yet: that scanner is written but switched off until it has been verified against an SCC Premium project. Exposure scans — buckets open to allUsers, Cloud SQL public IPs, unauthenticated Cloud Run services, public Pub/Sub topics — run every hour. IAM, Secret Manager and logging posture run once a night. An account administrator can ask for a fresh scan of a connection, at most once every 15 minutes per account; it runs with the next hourly scan, and while CloudMonitor is pre-launch nothing scans during the part of each day the service is powered down (see the Terms). Setup steps are on the Connectors page.